{ "id": 271732, "title": "nnAudio CQT1992v2 fully on GPU?", "url": "/competitions/g2net-gravitational-wave-detection/discussion/271732", "author_name": "", "post_date": "2021-09-12T09:24:54.197464900Z", "votes": 7, "comment_count": 5, "views": 0, "content": "

I am trying to run CQT1992v2 from the nnAudio library fully on GPU as a PyTorch layer.

\n

However, I am getting this error:

\n

RuntimeError: Input type (torch.cuda.FloatTensor) and weight type (torch.FloatTensor) should be the same

\n

After a quick check of the source code, it seems indeed that the self.cqt_kernels_realis on CPU.

\n

Has anyone managed to fix this? Maybe there is another method I am missing?

\n

If I find a solution, I will share it. Cheers!

\n

Update:

\n

Changing a bit the source code (by adding the device, here it is hardcoded), it seems to work:

\n

\"changes-cqt\"
jpg en html

", "messages": [ { "id": "1510305", "postDate": "09/12/2021 09:24:54", "content": "

I am trying to run CQT1992v2 from the nnAudio library fully on GPU as a PyTorch layer.

\n

However, I am getting this error:

\n

RuntimeError: Input type (torch.cuda.FloatTensor) and weight type (torch.FloatTensor) should be the same

\n

After a quick check of the source code, it seems indeed that the self.cqt_kernels_realis on CPU.

\n

Has anyone managed to fix this? Maybe there is another method I am missing?

\n

If I find a solution, I will share it. Cheers!

\n

Update:

\n

Changing a bit the source code (by adding the device, here it is hardcoded), it seems to work:

\n

\"changes-cqt\"
jpg en html

", "rawMarkdown": "I am trying to run [CQT1992v2](https://kinwaicheuk.github.io/nnAudio/_autosummary/nnAudio.Spectrogram.CQT1992v2.html#nnAudio.Spectrogram.CQT1992v2) from the nnAudio library fully on GPU as a PyTorch layer. \n\nHowever, I am getting this error: \n\n`RuntimeError: Input type (torch.cuda.FloatTensor) and weight type (torch.FloatTensor) should be the same`\n\nAfter a quick check of the source code, it seems indeed that the `self.cqt_kernels_real `is on CPU.\n\nHas anyone managed to fix this? Maybe there is another method I am missing?\n\nIf I find a solution, I will share it. Cheers!\n\n\n**Update**: \n\nChanging a bit the source code (by adding the device, here it is hardcoded), it seems to work:\n\n\n\"changes-cqt\"
jpg en html
", "votes": null }, { "id": "1510349", "postDate": "09/12/2021 10:20:11", "content": "

The error says that your cqt weights are not on cuda. Move your cqt to cuda then.

", "rawMarkdown": "The error says that your cqt weights are not on cuda. Move your cqt to cuda then.", "votes": null }, { "id": "1510358", "postDate": "09/12/2021 10:30:52", "content": "

Yes that's what I did as mentioned in the update however I was thinking if someone found a way to do it without changing the source code.

", "rawMarkdown": "Yes that's what I did as mentioned in the update however I was thinking if someone found a way to do it without changing the source code.", "votes": null }, { "id": "1510359", "postDate": "09/12/2021 10:32:28", "content": "

Just use nnAudio.Spectrogram.CQT1992v2(**params).cuda() in your create_cqt_kernels function, since CQT1992v2 inherits from torch.nn.Module, you can move this layer just like any built-in Conv2d / Linear layer

", "rawMarkdown": "Just use nnAudio.Spectrogram.CQT1992v2(**params).cuda() in your create_cqt_kernels function, since CQT1992v2 inherits from torch.nn.Module, you can move this layer just like any built-in Conv2d / Linear layer", "votes": null }, { "id": "1510360", "postDate": "09/12/2021 10:35:48", "content": "

Oh I didn't know this. Maybe that's what @cpmpml had in mind as well. Will give it a try. Thanks.

", "rawMarkdown": "Oh I didn't know this. Maybe that's what @cpmpml had in mind as well. Will give it a try. Thanks.", "votes": null }, { "id": "1559701", "postDate": "10/27/2021 07:07:41", "content": "

Hey All,

\n

Thank you all for taking part in our competition. The participation has been overwhelmingly positive. We are currently conducting a survey to gauge the demographic and outreach achieved. Kindly spare 2min and fill in this survey https://forms.gle/QP9L16niPexozyhu5.

\n

Thank you all,

\n

Regards,
\nChris

", "rawMarkdown": "Hey All,\n\nThank you all for taking part in our competition. The participation has been overwhelmingly positive. We are currently conducting a survey to gauge the demographic and outreach achieved. Kindly spare 2min and fill in this survey https://forms.gle/QP9L16niPexozyhu5.\n\nThank you all,\n\nRegards,\nChris", "votes": null } ], "comments": [ { "id": 1510349, "author_name": "cpmpml", "author_url": "", "post_date": "09/12/2021 10:20:11", "content": "

The error says that your cqt weights are not on cuda. Move your cqt to cuda then.

", "votes": null, "replies": [ { "id": 1510358, "author_name": "yassinealouini", "author_url": "", "post_date": "09/12/2021 10:30:52", "content": "

Yes that's what I did as mentioned in the update however I was thinking if someone found a way to do it without changing the source code.

", "votes": null, "replies": [] }, { "id": 1510359, "author_name": "martynoveduard", "author_url": "", "post_date": "09/12/2021 10:32:28", "content": "

Just use nnAudio.Spectrogram.CQT1992v2(**params).cuda() in your create_cqt_kernels function, since CQT1992v2 inherits from torch.nn.Module, you can move this layer just like any built-in Conv2d / Linear layer

", "votes": null, "replies": [] }, { "id": 1510360, "author_name": "yassinealouini", "author_url": "", "post_date": "09/12/2021 10:35:48", "content": "

Oh I didn't know this. Maybe that's what @cpmpml had in mind as well. Will give it a try. Thanks.

", "votes": null, "replies": [] } ] }, { "id": 1559701, "author_name": "zerafachris", "author_url": "", "post_date": "10/27/2021 07:07:41", "content": "

Hey All,

\n

Thank you all for taking part in our competition. The participation has been overwhelmingly positive. We are currently conducting a survey to gauge the demographic and outreach achieved. Kindly spare 2min and fill in this survey https://forms.gle/QP9L16niPexozyhu5.

\n

Thank you all,

\n

Regards,
\nChris

", "votes": null, "replies": [] } ], "raw_markdown_by_id": { "1510305": "I am trying to run [CQT1992v2](https://kinwaicheuk.github.io/nnAudio/_autosummary/nnAudio.Spectrogram.CQT1992v2.html#nnAudio.Spectrogram.CQT1992v2) from the nnAudio library fully on GPU as a PyTorch layer. \n\nHowever, I am getting this error: \n\n`RuntimeError: Input type (torch.cuda.FloatTensor) and weight type (torch.FloatTensor) should be the same`\n\nAfter a quick check of the source code, it seems indeed that the `self.cqt_kernels_real `is on CPU.\n\nHas anyone managed to fix this? Maybe there is another method I am missing?\n\nIf I find a solution, I will share it. Cheers!\n\n\n**Update**: \n\nChanging a bit the source code (by adding the device, here it is hardcoded), it seems to work:\n\n\n\"changes-cqt\"
jpg en html
", "1510349": "The error says that your cqt weights are not on cuda. Move your cqt to cuda then.", "1510358": "Yes that's what I did as mentioned in the update however I was thinking if someone found a way to do it without changing the source code.", "1510359": "Just use nnAudio.Spectrogram.CQT1992v2(**params).cuda() in your create_cqt_kernels function, since CQT1992v2 inherits from torch.nn.Module, you can move this layer just like any built-in Conv2d / Linear layer", "1510360": "Oh I didn't know this. Maybe that's what @cpmpml had in mind as well. Will give it a try. Thanks.", "1559701": "Hey All,\n\nThank you all for taking part in our competition. The participation has been overwhelmingly positive. We are currently conducting a survey to gauge the demographic and outreach achieved. Kindly spare 2min and fill in this survey https://forms.gle/QP9L16niPexozyhu5.\n\nThank you all,\n\nRegards,\nChris" }, "source": "meta" }